[llvm] r188879 - MC: ObjectSymbolizer can now recognize external function stubs.

Ahmed Bougacha ahmed.bougacha at gmail.com
Wed Aug 21 00:28:13 PDT 2013


Author: ab
Date: Wed Aug 21 02:28:13 2013
New Revision: 188879

URL: http://llvm.org/viewvc/llvm-project?rev=188879&view=rev
Log:
MC: ObjectSymbolizer can now recognize external function stubs.

Only implemented in the Mach-O ObjectSymbolizer.
The testcase sadly introduces a new binary.

Added:
    llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64   (with props)
Modified:
    llvm/trunk/include/llvm/MC/MCObjectSymbolizer.h
    llvm/trunk/lib/MC/MCObjectSymbolizer.cpp
    llvm/trunk/test/Object/X86/objdump-disassembly-symbolic.test

Modified: llvm/trunk/include/llvm/MC/MCObjectSymbolizer.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/MC/MCObjectSymbolizer.h?rev=188879&r1=188878&r2=188879&view=diff
==============================================================================
--- llvm/trunk/include/llvm/MC/MCObjectSymbolizer.h (original)
+++ llvm/trunk/include/llvm/MC/MCObjectSymbolizer.h Wed Aug 21 02:28:13 2013
@@ -56,6 +56,11 @@ public:
                                        int64_t Value, uint64_t Address);
   /// @}
 
+  /// \brief Look for an external function symbol at \p Addr.
+  /// (References through the ELF PLT, Mach-O stubs, and similar).
+  /// \returns An MCExpr representing the external symbol, or 0 if not found.
+  virtual StringRef findExternalFunctionAt(uint64_t Addr);
+
   /// \brief Create an object symbolizer for \p Obj.
   static MCObjectSymbolizer *
     createObjectSymbolizer(MCContext &Ctx, OwningPtr<MCRelocationInfo> &RelInfo,

Modified: llvm/trunk/lib/MC/MCObjectSymbolizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/MC/MCObjectSymbolizer.cpp?rev=188879&r1=188878&r2=188879&view=diff
==============================================================================
--- llvm/trunk/lib/MC/MCObjectSymbolizer.cpp (original)
+++ llvm/trunk/lib/MC/MCObjectSymbolizer.cpp Wed Aug 21 02:28:13 2013
@@ -26,9 +26,18 @@ using namespace object;
 
 namespace {
 class MCMachObjectSymbolizer : public MCObjectSymbolizer {
+  const MachOObjectFile *MOOF;
+  // __TEXT;__stubs support.
+  uint64_t StubsStart;
+  uint64_t StubsCount;
+  uint64_t StubSize;
+  uint64_t StubsIndSymIndex;
+
 public:
   MCMachObjectSymbolizer(MCContext &Ctx, OwningPtr<MCRelocationInfo> &RelInfo,
-                         const MachOObjectFile *MOOF) {}
+                         const MachOObjectFile *MOOF);
+
+  StringRef findExternalFunctionAt(uint64_t Addr) LLVM_OVERRIDE;
 
   void tryAddingPcLoadReferenceComment(raw_ostream &cStream,
                                        int64_t Value,
@@ -37,6 +46,62 @@ public:
 } // End unnamed namespace
 
 
+MCMachObjectSymbolizer::
+MCMachObjectSymbolizer(MCContext &Ctx, OwningPtr<MCRelocationInfo> &RelInfo,
+                       const MachOObjectFile *MOOF)
+    : MCObjectSymbolizer(Ctx, RelInfo, MOOF), MOOF(MOOF),
+      StubsStart(0), StubsCount(0), StubSize(0), StubsIndSymIndex(0) {
+
+  error_code ec;
+  for (section_iterator SI = MOOF->begin_sections(), SE = MOOF->end_sections();
+       SI != SE; SI.increment(ec)) {
+    if (ec) break;
+    StringRef Name; SI->getName(Name);
+    if (Name == "__stubs") {
+      SectionRef StubsSec = *SI;
+      if (MOOF->is64Bit()) {
+        macho::Section64 S = MOOF->getSection64(StubsSec.getRawDataRefImpl());
+        StubsIndSymIndex = S.Reserved1;
+        StubSize = S.Reserved2;
+      } else {
+        macho::Section S = MOOF->getSection(StubsSec.getRawDataRefImpl());
+        StubsIndSymIndex = S.Reserved1;
+        StubSize = S.Reserved2;
+      }
+      assert(StubSize && "Mach-O stub entry size can't be zero!");
+      StubsSec.getAddress(StubsStart);
+      StubsSec.getSize(StubsCount);
+      StubsCount /= StubSize;
+    }
+  }
+}
+
+StringRef MCMachObjectSymbolizer::findExternalFunctionAt(uint64_t Addr) {
+  // FIXME: also, this can all be done at the very beginning, by iterating over
+  // all stubs and creating the calls to outside functions. Is it worth it
+  // though?
+  if (!StubSize)
+    return StringRef();
+  uint64_t StubIdx = (Addr - StubsStart) / StubSize;
+  if (StubIdx >= StubsCount)
+    return StringRef();
+
+  macho::IndirectSymbolTableEntry ISTE =
+    MOOF->getIndirectSymbolTableEntry(MOOF->getDysymtabLoadCommand(), StubIdx);
+  uint32_t SymtabIdx = ISTE.Index;
+
+  StringRef SymName;
+  symbol_iterator SI = MOOF->begin_symbols();
+  error_code ec;
+  for (uint32_t i = 0; i != SymtabIdx; ++i) {
+    SI.increment(ec);
+  }
+  SI->getName(SymName);
+  assert(SI != MOOF->end_symbols() && "Stub wasn't found in the symbol table!");
+  assert(SymName.front() == '_' && "Mach-O symbol doesn't start with '_'!");
+  return SymName.substr(1);
+}
+
 void MCMachObjectSymbolizer::
 tryAddingPcLoadReferenceComment(raw_ostream &cStream, int64_t Value,
                                 uint64_t Address) {
@@ -71,6 +136,16 @@ bool MCObjectSymbolizer::
 tryAddingSymbolicOperand(MCInst &MI, raw_ostream &cStream,
                          int64_t Value, uint64_t Address, bool IsBranch,
                          uint64_t Offset, uint64_t InstSize) {
+  if (IsBranch) {
+    StringRef ExtFnName = findExternalFunctionAt((uint64_t)Value);
+    if (!ExtFnName.empty()) {
+      MCSymbol *Sym = Ctx.GetOrCreateSymbol(ExtFnName);
+      const MCExpr *Expr = MCSymbolRefExpr::Create(Sym, Ctx);
+      MI.addOperand(MCOperand::CreateExpr(Expr));
+      return true;
+    }
+  }
+
   if (const RelocationRef *R = findRelocationAt(Address + Offset)) {
     if (const MCExpr *RelExpr = RelInfo->createExprForRelocation(*R)) {
       MI.addOperand(MCOperand::CreateExpr(RelExpr));

Added: llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64?rev=188879&view=auto
==============================================================================
Binary files llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64 (added) and llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64 Wed Aug 21 02:28:13 2013 differ

Propchange: llvm/trunk/test/Object/Inputs/trivial-executable-test.macho-x86-64
------------------------------------------------------------------------------
    svn:executable = *

Modified: llvm/trunk/test/Object/X86/objdump-disassembly-symbolic.test
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Object/X86/objdump-disassembly-symbolic.test?rev=188879&r1=188878&r2=188879&view=diff
==============================================================================
--- llvm/trunk/test/Object/X86/objdump-disassembly-symbolic.test (original)
+++ llvm/trunk/test/Object/X86/objdump-disassembly-symbolic.test Wed Aug 21 02:28:13 2013
@@ -3,6 +3,11 @@ RUN:              | FileCheck %s -check-
 RUN: llvm-objdump -d -symbolize %p/../Inputs/trivial-object-test.macho-x86-64 \
 RUN:              | FileCheck %s -check-prefix MACHO-x86-64
 
+# Generate this using:
+#   ld trivial-object-test.macho-x86-64 -undefined dynamic_lookup
+RUN: llvm-objdump -d -symbolize %p/../Inputs/trivial-executable-test.macho-x86-64 \
+RUN:              | FileCheck %s -check-prefix MACHO-STUBS-x86-64
+
 ELF-x86-64: file format ELF64-x86-64
 ELF-x86-64: Disassembly of section .text:
 ELF-x86-64: main:
@@ -28,3 +33,16 @@ MACHO-x86-64:       1a:	e8 00 00 00 00
 MACHO-x86-64:       1f:	8b 44 24 04                                  	movl	4(%rsp), %eax
 MACHO-x86-64:       23:	48 83 c4 08                                  	addq	$8, %rsp
 MACHO-x86-64:       27:	c3                                           	ret
+
+MACHO-STUBS-x86-64: file format Mach-O 64-bit x86-64
+MACHO-STUBS-x86-64: Disassembly of section __TEXT,__text:
+MACHO-STUBS-x86-64: _main:
+MACHO-STUBS-x86-64:     1f90:       48 83 ec 08                                     subq    $8, %rsp
+MACHO-STUBS-x86-64:     1f94:       c7 44 24 04 00 00 00 00                         movl    $0, 4(%rsp)
+MACHO-STUBS-x86-64:     1f9c:       48 8d 3d 45 00 00 00                            leaq    69(%rip), %rdi ## literal pool for: Hello World!
+MACHO-STUBS-x86-64:     1fa3:       e8 16 00 00 00                                  callq   puts
+MACHO-STUBS-x86-64:     1fa8:       30 c0                                           xorb    %al, %al
+MACHO-STUBS-x86-64:     1faa:       e8 09 00 00 00                                  callq   SomeOtherFunction
+MACHO-STUBS-x86-64:     1faf:       8b 44 24 04                                     movl    4(%rsp), %eax
+MACHO-STUBS-x86-64:     1fb3:       48 83 c4 08                                     addq    $8, %rsp
+MACHO-STUBS-x86-64:     1fb7:       c3                                              ret





More information about the llvm-commits mailing list